Desenvolvimento de um aplicativo para o auxílio no diagnóstico de perda óssea radiográfica periodontal via aprendizagem profunda

Abstract

Periodontitis is a multifactorial chronic inflammatory disease associated with an accumulation of dysbiotic biofilm and characterized by progressive bone loss, which can be measured by the distance between the Cemento-Enamel Junction (CEJ) and the alveolar bone crest (ABC) on radiographs, known as radiographic bone loss (RBL). The use of Convolutional Neural Networks (CNNs) has shown promise in the analysis and interpretation of dental radiograph images, making it faster and more objective. Therefore, the study aimed to develop an application for the detection and measurement of periodontal RBL using CNN. This study included 595 digital radiographic images of interproximal type, which underwent keypoint annotations using the CVAT platform. Out of these, 565 images were trained using the YOLOv8-pose architecture, resulting in a model for detecting keypoints related to the CEJ and ABC. Based on the obtained model, a web application was developed using the Streamlit tool, capable of detecting and measuring RBL by calculating the Euclidean distance between the points. The model achieved promising results, with a precision of 0.611, a recall of 0,739, and an mAP of 73,8%. The application demonstrated good functionality and performance, accessible on both computers and smartphones. In conclusion, the specific model and application for identifying and measuring radiographic bone loss showed remarkable performance and robust average accuracy in diagnosing this periodontal indicator. Therefore, the potential for automating the diagnostic process through this innovative web application is significant.